On Mon, 9 Sep 2024 12:57:17 GMT, Maurizio Cimadamore <mcimadam...@openjdk.org> wrote:
> The new test added by https://github.com/openjdk/jdk/pull/20854 fails > spuriously. > While JNI lookup is now moved into the static initializer of the > `MappedMemoryUtils` class, this class might only get initialized while in the > middle of a scoped context. > > To address this, I created a new proxy interface, namely > `MappedMemoryUtilsProxy`. This interface is used by `ScopedMemoryAccess` to > call the various `force`/`load`/`isLoaded`/`unload` methods, and a singleton > instance is provided inside `MappedMemoryUtils` itself, and then exposed via > the `SharedSecrets` mechanism. > > Crucially, `MappedMemorySegmentImpl` will now _first_ obtain a > `MappedMemoryUtilsProxy` and _then_ call `ScopedMemoryAccess`. This should > move all class initializer side-effects out of the scoped method context. src/java.base/share/classes/jdk/internal/misc/X-ScopedMemoryAccess.java.template line 255: > 253: session.checkValidStateRaw(); > 254: } > 255: return mappedUtils.isLoaded(address, isSync, size); this method (and other related methods) now only performs a straight call into `MappedMemoryUtilsProxy`. Static initializers and JNI resolution has already run, so we should no longer use too much stack. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/20914#discussion_r1750259314
